Work with your agent
Use Claude or ChatGPT/Codex to investigate a question. Biosimulant keeps the plans, Lab revisions, runs, results and evidence saved through its services together, so you can inspect the work, repeat a setup and continue it later.
Your agent does the work in the assistant you already use. Studio is where you follow and review the saved record. You can begin with a research question; you do not need a model or a run to start a session.
Start here
| What you want to do | Guide |
|---|---|
| Connect your assistant to your Biosimulant account | Connect your agent |
| Start a session and follow the work in Studio | Start a recorded investigation |
| Understand what is saved and continue existing work | Sessions and the research record |
| Review changes, runs and publication before they happen | Approvals and grants |
| Use the tools directly or write an integration | Agent tool reference |
Connecting uses MCP, a standard way for an assistant to use an outside service. Sign in through your assistant’s browser setup. You do not need to install the Biosimulant runtime or CLI.
Keep the investigation together
Connect your agent
Connect the assistant you use, then start a session in Studio. Copy the session prompt into your assistant. It tells the agent which session to use and what you want to investigate.
You can also ask a connected agent to find an existing session or start one. Reuse the session when continuing the same investigation, so its saved work stays together.
Inspect what ran
Open the session’s Timeline to see recorded plans, Lab changes, runs and saved documents. Results brings the outputs together; Evidence shows saved sources, checks and reviews.
Open a run to inspect its inputs, settings, results and log. Use Ask in Studio to ask about the saved record without starting a run or changing the Lab.
Rerun the saved setup
Ask your agent to repeat a saved run, using the same Lab revision, inputs and settings. Review its proposed setup before approving the run. If you want to change something, ask it to identify the change so you can compare the outcomes.
A saved setup makes a run repeatable, but models with randomness can still produce different numbers.
Build on previous work
Reopen a session from Sessions in Studio. Select Continue in your agent, copy the prompt and give it to the assistant you want to use next. It can read the saved state and work out what remains.
For shared work, each teammate can connect their own agent to the team workspace. The saved work belongs to the team. Access and approval still depend on each person’s permissions.
What to ask for
Start with a goal and say what you want kept in the session. For example:
Continue this session. Review the saved evidence, explain what we know and what is missing, and propose the next step. Do not run or change anything yet.
Inspect the completed run in this session. Explain its inputs, settings, results and limitations using the saved record.
Repeat that run with one input changed. Show me the difference in the setup before asking me to approve it.
Continue the investigation from the approved plan. Check what has already been done before proposing more work.
These are example requests, not promises that an agent will always produce a particular answer. Review the saved record and any gaps it reports.
Models, experiments and publishing
When the question needs computation, your agent can find a published Lab, create a private copy, build or adapt its models, and prepare a run. A Lab is the executable setup; the session keeps the investigation around it.
- Use experiments to compare several setups against one question.
- Use model qualification when you need explicit requirements and acceptance checks.
- Read Passports and evidence before deciding what a result supports.
- Review publication and visibility before sharing a release.
What gets recorded
Biosimulant records work performed through its services and material saved to the session or Lab workspace. Connecting an agent does not copy its whole conversation, every search or work done through other tools into Studio.
Ask the agent to save supporting material using the available file and evidence tools, then check that you can find it. A successful run or a finished plan does not by itself establish that the result is scientifically valid.
See Sessions and the research record for the details.